main.py

from dotenv import load_dotenv
import os

from langchain_core.messages import HumanMessage
from langchain_core.runnables.config import RunnableConfig

from config import load_config
from tools import create_tools
from graph import create_agent_graph

import chainlit as cl
import re

# Initialize resources at module level for Chainlit
load_dotenv()
config = load_config()
tools, model = create_tools(config)
graph = create_agent_graph(tools, model)

def clean_content(content):
    """Clean and format content for better readability.
    
    Args:
        content: The content to clean
        
    Returns:
        Cleaned content string
    """
    if not content:
        return ""
    
    # Replace escaped newlines with actual newlines
    content = content.replace('\\n', '\n')
    
    # Remove excessive backslashes
    content = content.replace('\\\\', '\\')
    
    # Format URLs properly
    content = re.sub(r'(?<![\[\("])https?://[^\s\'"]+', r'[Link](\g<0>)', content)
    
    # Remove raw markdown links that didn't render
    content = re.sub(r'\\*\[([^\]]+)\]\\*\(([^)]+)\)', r'[\1](\2)', content)
    
    # Clean up repeated newlines
    content = re.sub(r'\n{3,}', '\n\n', content)
    
    return content

@cl.on_chat_start
async def on_chat_start():
    """Set up the session state when a new chat starts."""
    cl.user_session.set("graph", graph)
    # Send welcome message
    welcome = cl.Message(content="👋 Hello! I'm a research assistant powered by LangGraph. I can search the web using Tavily and query arXiv papers. How can I help you today?")
    await welcome.send()

@cl.on_message
async def on_message(message: cl.Message):
    """Process incoming user messages with the LangGraph agent."""
    try:
        # Get the compiled graph from session
        graph = cl.user_session.get("graph")
        
        # Create input with the user's message
        inputs = {
            "messages": [
                HumanMessage(content=message.content)
            ]
        }
        
        # Set up callback handler for streaming - IMPORTANT CHANGE HERE
        cb = cl.LangchainCallbackHandler(stream_final_answer=True)
        runnable_config = RunnableConfig(callbacks=[cb])
        
        # Send thinking indicator
        thinking = cl.Message(content="🤔 Thinking...")
        await thinking.send()
        
        # Process with agent using invoke instead of stream to ensure completion
        result = await graph.ainvoke(inputs, config=runnable_config)
        
        # Remove thinking indicator once we get results
        await thinking.remove()
        
        # If no response was streamed, send the final response explicitly
        if "messages" in result and result["messages"]:
            final_message = result["messages"][-1]
            if hasattr(final_message, "content") and final_message.content:
                content = clean_content(final_message.content)
                await cl.Message(content=content).send()
                
    except Exception as e:
        error_msg = f"An error occurred: {str(e)}"
        print(f"ERROR: {error_msg}")
        await cl.Message(content=f"❌ {error_msg}").send()
        
# Only used when running the script directly, not with Chainlit
if __name__ == "__main__":
    print("Please run this app with Chainlit: chainlit run main.py")